Implementación y validación del módulo basado en casos de un modelo híbrido
Archivos
Fecha
2005-07-02
Autores
Couce Sardiñas, Yasel
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad Central “Marta Abreu” de Las Villas
Resumen
En el grupo de Inteligencia Artificial se diseñó un modelo híbrido que combina Redes Neuronales Artificiales (RNA) y Razonamiento Basado en Casos (RBC), donde la RNA propone una solución a un problema y el RBC la justifica en el contexto de los casos más semejantes a este. El modelo de RNA implementado es asociativo, y por tanto en su topología se asocia cada valor presente para un rasgo en la base de casos a una neurona. En este sentido se extiende este modelo utilizando conjuntos borrosos, y así considerar los términos lingüísticos como los valores representativos en este tipo de rasgo.
Por otro lado, una muestra de aprendizaje para la RNA puede que no sea representativa del dominio del problema, y por tanto la RNA no lograría generalizaciones correctas. Otra situación pudiera ser que se presenten problemas, los cuales por sus características se considerarían excepciones del dominio y cuya solución por la RNA puede no ser correcta. En estos casos el enfoque híbrido del modelo original se pudiera extender, utilizando estos pesos de la RNA en una función de semejanza similar a la anterior, donde el RBC resuelva el problema.
En el presente trabajo se implementa una herramienta computacional para la componente basada en casos del nuevo modelo híbrido, en la cual el RBC desempeña los roles de justificador y resolvedor. La validación utilizando archivos de datos internacionales con la herramienta desarrollada, permite concluir la factibilidad de estas extensiones al modelo original.
A hybrid model that combines Artificial Neural Networks (ANN) and Case-Based Reasoning (CBR) was designed at the Artificial Intelligence research group. In it, the ANN proposes a solution to a problem and the CBR justifies it in the context of the most similar cases to that one. The ANN implemented model is an associative one, and therefore in its topology each feature’s value that appears in the cases base is associated with a neuron. In this sense the hybrid model is extended using fuzzy sets, considering linguistic terms as representative values for this type of features. On the other hand, a learning sample for the ANN might not be representative of the problem’s domain, and therefore the ANN would not obtain correct generalizations. Another situation could be that certain specific problems may appear -which could be considered by its characteristics as exceptions of the domain- and whose solution by the ANN could not be right. In these cases, the hybrid approach of the original model could be extended using these weights from ANN in a similarity function where the CBR solves the problem. In this present work, a computational tool for the cases-based component of the new hybrid model is implemented where the CBR carries out both the rolls of justifier and solver. The validation using well-known international data files with the developed tool allows concluding the feasibility of these extensions to the original model.
A hybrid model that combines Artificial Neural Networks (ANN) and Case-Based Reasoning (CBR) was designed at the Artificial Intelligence research group. In it, the ANN proposes a solution to a problem and the CBR justifies it in the context of the most similar cases to that one. The ANN implemented model is an associative one, and therefore in its topology each feature’s value that appears in the cases base is associated with a neuron. In this sense the hybrid model is extended using fuzzy sets, considering linguistic terms as representative values for this type of features. On the other hand, a learning sample for the ANN might not be representative of the problem’s domain, and therefore the ANN would not obtain correct generalizations. Another situation could be that certain specific problems may appear -which could be considered by its characteristics as exceptions of the domain- and whose solution by the ANN could not be right. In these cases, the hybrid approach of the original model could be extended using these weights from ANN in a similarity function where the CBR solves the problem. In this present work, a computational tool for the cases-based component of the new hybrid model is implemented where the CBR carries out both the rolls of justifier and solver. The validation using well-known international data files with the developed tool allows concluding the feasibility of these extensions to the original model.
Descripción
Palabras clave
Redes Neuronales Artificiales (RNA), Razonamiento Basado en Casos (RBC), Conjuntos Borrosos, Modelo Híbrido, Diseño, Inteligencia Artificial